A server cluster anomaly diagnosis method based on big data AI

Abstract

The present invention discloses a server cluster abnormality diagnosis method based on big data AI, which relates to the field of data mining technology, including: quantifying the component maintenance health growth coefficient of the server cluster; evaluating the component initialization health index of the server cluster; correcting the component initialization health index of the server cluster according to the component maintenance health growth coefficient of the server cluster to obtain the real-time component health status of the server cluster; collecting the real-time operation data of the server cluster components; constructing the abnormal event operation parameter classification model of the server cluster components, generating the abnormal event data distribution decision boundary diagram of the server cluster components; correlating and matching the abnormal event data distribution decision boundary diagram of the server cluster components with the real-time operation data of the server cluster components, and generating the abnormal event development direction of the server cluster components. The advantages of the present invention are: real-time diagnosis of the abnormality of the server cluster and improving the maintenance efficiency of the server cluster.

Description

Technical Field

[0001] The present invention relates to the technical field of data mining, and in particular to a server cluster anomaly diagnosis method based on big data AI. Background Art

[0002] Server cluster anomaly diagnosis refers to timely discovering and locating anomalies or faults in the cluster by monitoring and analyzing the operating data and health status of each component in the server cluster, so as to quickly take repair measures to ensure the stability of the cluster and business continuity.

[0003] The existing server cluster anomaly diagnosis will only issue an alarm when an abnormality occurs in the monitoring component target. That is, the abnormal event can only be effectively detected and diagnosed after it has significantly affected system performance or caused service interruption. This leads to a lag in the identification of server cluster anomalies, which reduces the operating stability of the overall equipment and increases business risks and maintenance costs. Summary of the invention

[0004] In order to solve the above technical problems, a server cluster anomaly diagnosis method based on big data AI is provided. This technical solution solves the above-mentioned problem that the existing server cluster anomaly diagnosis will only issue an alarm when an abnormality occurs in the monitoring component target, that is, the abnormal event can only be effectively detected and diagnosed after it has significantly affected the system performance or caused service interruption, which makes the server cluster anomaly identification have a lag, reduces the operating stability of the overall equipment, and increases business risks and maintenance costs.

[0005] In order to achieve the above purpose, the technical solution adopted by the present invention is:

[0006] A server cluster abnormality diagnosis method based on big data AI, comprising:

[0007] Based on the historical maintenance logs of the server cluster, the health growth value of each component of the server cluster due to the historical maintenance task is quantified to obtain the component maintenance health growth coefficient of the server cluster.

[0008] Based on the historical operation logs of the server cluster, analyze the deviation trend of the historical operation data of each component of the server cluster and evaluate the component initialization health indicators of the server cluster;

[0009] According to the component maintenance health growth coefficient of the server cluster, the component initialization health index of the server cluster is corrected to evaluate the real-time component health status of the server cluster;

[0010] Collect real-time operation data of components in the server cluster according to the real-time component health status of the server cluster;

[0011] Capture abnormal events of server cluster components based on the Internet, obtain operational characteristic parameters of abnormal events of server cluster components, and construct an operational characteristic parameter matrix of abnormal events of server cluster components;

[0012] Based on the abnormal event operation characteristic parameter matrix of the server cluster component, a classification model of the abnormal event operation parameters of the server cluster component is constructed, and a decision boundary diagram of the abnormal event data distribution of the server cluster component is generated;

[0013] Based on the abnormal event data distribution decision boundary diagram of the server cluster components and the real-time operation data of the server cluster components, the development direction of the abnormal events of the server cluster components is generated.

[0014] Preferably, based on the historical maintenance log of the server cluster, the health growth value of each component of the server cluster for the historical maintenance task is quantified, and the component maintenance health growth coefficient of the server cluster is evaluated, which specifically includes:

[0015] Based on the historical maintenance log of the server cluster, extract the historical maintenance record of the server cluster component in unit time; the historical maintenance record includes: maintenance time, maintenance type, and maintenance component;

[0016] Divide the historical maintenance records of the server cluster components according to unit time to obtain a time series data array of the maintenance records of the server cluster components;

[0017] According to the maintenance record time series data of the server cluster component, the performance value of the maintenance component corresponding to the start time node of the maintenance task is used as the initial health benchmark value, and the maintenance type of the maintenance task for the health growth value of the maintenance component is calculated;

[0018] Count the average growth value of the health growth value of the maintenance component for all maintenance types of the maintenance tasks in the maintenance record time series data of the server cluster component;

[0019] Calculate the component maintenance health growth coefficient of the server cluster based on the health growth value of the maintenance type of the maintenance task for the maintenance component and the average growth value of the health growth value of the maintenance type of the maintenance task for the maintenance component;

[0020] The component maintenance health growth coefficient of the server cluster is specifically:

[0021]

[0022] Where Hup i Maintain a healthy growth coefficient, ΔH, for the i-th component of the server cluster ij is the health growth value of the i-th component of the server cluster after the j-th maintenance task, is the average health growth value of the i-th component of the server cluster, is the average health growth value of all components in the server cluster, n i is the total maintenance times of the i-th component, are the performance score of the i-th component of the server cluster after the j-th maintenance task, and the performance score of the i-th component of the server cluster before the j-th maintenance task, respectively. are the maximum performance score of the i-th component of the server cluster in the j-th maintenance task, the minimum performance score of the i-th component of the server cluster in the j-th maintenance task, and W ij The weight of the maintenance task for the i-th component of the server cluster at the jth time;

[0023] Preferably, based on the historical operation log of the server cluster, the deviation trend of the historical operation data of each component of the server cluster is analyzed, and the component initialization health index of the server cluster is evaluated, which specifically includes:

[0024] Based on the historical operation logs of the server cluster, the operation data of the server cluster components are associated according to the time attributes of the historical operation logs to create the server cluster component operation sequence data;

[0025] Preprocessing is performed based on the server cluster component runtime data, and the operation characteristic parameters of each component in the server cluster component runtime data are counted, and an operation deviation threshold T is set; the abnormal operation deviation threshold is a preset constant, which is obtained based on the normal distribution Z score of the data;

[0026] The time span between the maintenance time nodes of the historical maintenance tasks is used as the sliding window length, and the abnormal operation characteristic parameters of each component operation characteristic parameter within the sliding window length that exceeds the operation deviation threshold are recursively filtered out. Among them, e′ i is the abnormal operation characteristic parameter of the i-th component, e i is the operating characteristic parameter of the i-th component, is the mean value of the operating characteristic parameter, S is the standard deviation of the operating characteristic parameter;

[0027] Based on the abnormal operation characteristic parameters of each component of the server cluster within the sliding window length, the abnormal deviation degree and abnormal deviation time duration under the corresponding abnormal state are marked, and the abnormal operation deviation weight of each component is calculated; the abnormal deviation degree refers to the absolute value between the abnormal operation characteristic parameter and the normal operation characteristic parameter;

[0028] Based on the abnormal operation characteristic parameters of each component of the server cluster and the abnormal operation deviation weight of each component, the component initialization health index of the server cluster is calculated;

[0029] The component initialization health indicators of the server cluster are specifically:

[0030]

[0031] In the formula, G i Initialize the health indicator, W′, for the i-th component of the server cluster ik is the kth abnormal operation feature parameter weight of the i-th component of the server cluster, D ik is the deviation value of the kth abnormal operation characteristic parameter of the i-th component of the server cluster, U ik is the deviation time of the kth abnormal operation characteristic parameter of the i-th component of the server cluster, W′ max is the maximum weight of the abnormal operation characteristic parameter of the server cluster component within the sliding window length, and m is the total number of abnormal operation characteristic parameters.

[0032] Preferably, based on the abnormal event operation parameter matrix of the server cluster component, constructing the abnormal event operation parameter classification model of the server cluster component, and generating the abnormal event data distribution decision boundary diagram of the server cluster component specifically includes:

[0033] Based on the principal component analysis method, the dimension reduction process is performed on each element in the abnormal event operation parameter matrix of the server cluster component to obtain the reduced dimension server cluster component abnormal event operation parameter matrix;

[0034] Based on SVM vector machine, a classification model of abnormal events of server cluster components is constructed;

[0035] Based on the dimension-reduced server cluster component abnormal event operation parameter matrix, the dimension-reduced operation parameters of each component abnormal event are used to construct a binary scatter plot of the server cluster component abnormal event operation parameters;

[0036] Substitute the binary scatter plot of the abnormal event operation parameters of the server cluster component into the abnormal event classification model of the server cluster component, perform cyclic iteration to divide the optimal distance between each scatter point in the binary scatter plot and the decision surface, take the minimization error function of the decision boundary graph of the model training as the end goal, and generate the abnormal event data distribution decision boundary graph of the server cluster component;

[0037] The server cluster component abnormal event classification model is specifically:

[0038]

[0039] In the formula, F ij is the decision boundary diagram of abnormal event data distribution of server cluster components, w * is the optimal weight vector, b * is the optimal bias term, is the optimal value of the Lagrange multiplier associated with the kth abnormal operation characteristic parameter in the binary scatter plot, y k is the true label of the kth abnormal operation feature parameter in the binary scatter plot, x k is the kth abnormal operation characteristic parameter in the binary scatter plot, y c is the true label of the cth support vector, x c is the eigenvector of the c-th support vector.

[0040] Preferably, the abnormal event data distribution decision boundary diagram of the server cluster component is associated and matched with the real-time operation data of the server cluster component to generate the development direction of the abnormal event of the server cluster component, which specifically includes:

[0041] Standardize the real-time running data of the components of the server cluster, associate it with the real-time running data of the components of the server cluster according to the unit time, and create the real-time running sequence data of the components of the server cluster;

[0042] Normalize the real-time component health status values ​​of the server cluster and assign health status weights to the corresponding components of the server cluster.

[0043] Based on the component health status weight of the server cluster, the real-time running time data of the server cluster components, and the abnormal event data distribution decision boundary diagram of the server cluster components, the initial server cluster component abnormal development direction matching value is obtained;

[0044] Taking the unit time as the recursive step length, the matching growth value of the abnormal development direction of the initial server cluster components under the cumulative recursive step length is analyzed;

[0045] Determine whether the matching growth value of the initial server cluster component abnormal development direction under the cumulative recursive step length is within the abnormal event matching threshold range of the server cluster component. If not, continue to observe. If so, issue a corresponding server cluster component abnormal event.

[0046] The specific matching value of the abnormal development direction of the initial server cluster component is:

[0047]

[0048] In the formula, cos i is the abnormal event matching value of the i-th component of the server cluster, Δcos i Match the growth value of the abnormal event of the i-th component of the server cluster, cos i (t+1) is the matching growth value of the abnormal development direction of the initial server cluster component under the cumulative t+1th recursive step, Q iis the real-time operating parameter of the i-th component of the server cluster, w i is the health status weight of the i-th component of the server cluster.

[0049] Compared with the prior art, the present invention has the following beneficial effects:

[0050] The present invention proposes a server cluster abnormality diagnosis solution based on big data AI, which evaluates the component health status by quantifying the historical maintenance data growth value and operation data deviation trend of server cluster components, builds a classification model and generates a decision boundary graph in combination with abnormal event operation parameters, so as to match component operation data in real time and predict the development direction of abnormal components. The beneficial effect is that the abnormality of the server cluster can be diagnosed in real time, and the maintenance efficiency of the server cluster can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 This is a flow chart of a server cluster anomaly diagnosis method based on big data AI;

[0052] Figure 2 A flow chart of the method for evaluating the maintenance of healthy growth coefficients for components of a server cluster;

[0053] Figure 3 A flow chart of a method for initializing health indicators for evaluating components of a server cluster;

[0054] Figure 4 A flow chart of a method for generating a decision boundary graph of abnormal event data distribution of a server cluster component;

[0055] Figure 5 To generate a flow chart of the development direction of abnormal events of server cluster components. DETAILED DESCRIPTION

[0056] The following description is used to disclose the present invention so that those skilled in the art can implement the present invention. The preferred embodiments described below are only examples, and those skilled in the art may think of other obvious variations.

[0057] Reference Figure 1 As shown, a server cluster abnormality diagnosis method based on big data AI includes:

[0058] Based on the historical maintenance logs of the server cluster, the health growth value of each component of the server cluster due to the historical maintenance task is quantified to obtain the component maintenance health growth coefficient of the server cluster.

[0059] Based on the historical operation logs of the server cluster, analyze the deviation trend of the historical operation data of each component of the server cluster and evaluate the component initialization health indicators of the server cluster;

[0060] According to the component maintenance health growth coefficient of the server cluster, the component initialization health index of the server cluster is corrected to evaluate the real-time component health status of the server cluster;

[0061] Collect real-time operation data of components in the server cluster according to the real-time component health status of the server cluster;

[0062] Capture abnormal events of server cluster components based on the Internet, obtain operational characteristic parameters of abnormal events of server cluster components, and construct an operational characteristic parameter matrix of abnormal events of server cluster components;

[0063] Based on the abnormal event operation characteristic parameter matrix of the server cluster component, a classification model of the abnormal event operation parameters of the server cluster component is constructed, and a decision boundary diagram of the abnormal event data distribution of the server cluster component is generated;

[0064] Based on the abnormal event data distribution decision boundary diagram of the server cluster components and the real-time operation data of the server cluster components, the development direction of the abnormal events of the server cluster components is generated.

[0065] It is understandable that the processing core performance of the server cluster component is very slow to lose performance in the actual operation without human interference. The actual factor causing the performance degradation is that the auxiliary component ages after long-term operation, resulting in loss and congestion of communication information, which makes the server cluster abnormal. Therefore, the core of this solution is to analyze the status analysis of the auxiliary component to evaluate the similarity between the operating data under the corresponding status of different components and the abnormal events of the components, so as to achieve early intervention of abnormal diagnosis;

[0066] This solution evaluates the health status of components by quantifying the historical maintenance data growth value and operation data deviation trend of server cluster components, and builds a classification model and generates a decision boundary diagram based on the abnormal event operation parameters, so as to match the component operation data in real time and predict the development direction of abnormal components. The beneficial effect is that it can diagnose the abnormality of the server cluster in real time and improve the maintenance efficiency of the server cluster.

[0067] Reference Figure 2 As shown in the figure, based on the historical maintenance log of the server cluster, the health growth value of each component of the server cluster for the historical maintenance task is quantified, and the component maintenance health growth coefficient of the server cluster is evaluated, which specifically includes:

[0068] Based on the historical maintenance log of the server cluster, extract the historical maintenance record of the server cluster component in unit time; the historical maintenance record includes: maintenance time, maintenance type, and maintenance component;

[0069] Divide the historical maintenance records of the server cluster components according to unit time to obtain a time series data array of the maintenance records of the server cluster components;

[0070] According to the maintenance record time series data of the server cluster component, the performance value of the maintenance component corresponding to the start time node of the maintenance task is used as the initial health benchmark value, and the maintenance type of the maintenance task for the health growth value of the maintenance component is calculated;

[0071] Count the average growth value of the health growth value of the maintenance component for all maintenance types of the maintenance tasks in the maintenance record time series data of the server cluster component;

[0072] Calculate the component maintenance health growth coefficient of the server cluster based on the health growth value of the maintenance type of the maintenance task for the maintenance component and the average growth value of the health growth value of the maintenance type of the maintenance task for the maintenance component;

[0073] The component maintenance health growth coefficient of the server cluster is specifically:

[0074]

[0075] Where Hup i Maintain a healthy growth coefficient, ΔH, for the i-th component of the server cluster ij is the health growth value of the i-th component of the server cluster after the j-th maintenance task, is the average health growth value of the i-th component of the server cluster, is the average health growth value of all components in the server cluster, n i is the total maintenance times of the i-th component, are the performance score of the i-th component of the server cluster after the j-th maintenance task, and the performance score of the i-th component of the server cluster before the j-th maintenance task, respectively. are the maximum performance score of the i-th component of the server cluster in the j-th maintenance task, the minimum performance score of the i-th component of the server cluster in the j-th maintenance task, and W ij The jth maintenance task weight for the i-th component of the server cluster.

[0076] It should be noted that the weight of the maintenance task is determined by the type of maintenance task. For example, when the maintenance task of the server cluster is to maintain the PCB motherboard or the components therein, the corresponding weight distribution is higher, and for normal maintenance of the components, the weight distribution can be averaged:

[0077] It is understandable that due to differences in the capabilities of maintenance personnel, there are basic differences in the actual growth of the component health value of the server cluster for each maintenance task. In order to ensure the accuracy of the usage data analysis in subsequent steps, the component maintenance health growth coefficient of the server cluster is calculated by analyzing historical maintenance tasks to reduce the fine-grained performance fluctuations of the data.

[0078] Reference Figure 3 As shown in the figure, based on the historical operation logs of the server cluster, the deviation trend of the historical operation data of each component of the server cluster is analyzed, and the component initialization health indicators of the server cluster are evaluated, including:

[0079] Based on the historical operation logs of the server cluster, the operation data of the server cluster components are associated according to the time attributes of the historical operation logs to create the server cluster component operation sequence data;

[0080] Preprocessing is performed based on the server cluster component runtime data, and the operation characteristic parameters of each component in the server cluster component runtime data are counted, and an operation deviation threshold T is set; the abnormal operation deviation threshold is a preset constant, which is obtained based on the normal distribution Z score of the data;

[0081] The time span between the maintenance time nodes of the historical maintenance tasks is used as the sliding window length, and the abnormal operation characteristic parameters of each component operation characteristic parameter within the sliding window length that exceeds the operation deviation threshold are recursively filtered out. Among them, e′ i is the abnormal operation characteristic parameter of the i-th component, e i is the operating characteristic parameter of the i-th component, is the mean value of the operating characteristic parameter, S is the standard deviation of the operating characteristic parameter;

[0082] Based on the abnormal operation characteristic parameters of each component of the server cluster within the sliding window length, mark the abnormal deviation degree and abnormal deviation time duration under the corresponding abnormal state, and calculate the abnormal operation deviation weight of each component; the abnormal deviation degree refers to the absolute value between the abnormal operation characteristic parameter and the normal operation characteristic parameter;

[0083] Based on the abnormal operation characteristic parameters of each component of the server cluster and the abnormal operation deviation weight of each component, the component initialization health index of the server cluster is calculated;

[0084] The component initialization health indicators of the server cluster are specifically:

[0085]

[0086] In the formula, G iInitialize the health indicator, W′, for the i-th component of the server cluster ik is the kth abnormal operation feature parameter weight of the i-th component of the server cluster, D ik is the deviation value of the kth abnormal operation characteristic parameter of the i-th component of the server cluster, U ik is the deviation time of the kth abnormal operation characteristic parameter of the i-th component of the server cluster, W′ max is the maximum weight of the abnormal operation characteristic parameter of the server cluster component within the sliding window length, and m is the total number of abnormal operation characteristic parameters.

[0087] Reference Figure 4 As shown, based on the abnormal event operation parameter matrix of the server cluster component, the abnormal event operation parameter classification model of the server cluster component is constructed, and the abnormal event data distribution decision boundary diagram of the server cluster component is generated, which specifically includes:

[0088] Based on the principal component analysis method, the dimension reduction process is performed on each element in the abnormal event operation parameter matrix of the server cluster component to obtain the reduced dimension server cluster component abnormal event operation parameter matrix;

[0089] Based on SVM vector machine, a classification model of abnormal events of server cluster components is constructed;

[0090] Based on the dimension-reduced server cluster component abnormal event operation parameter matrix, the dimension-reduced operation parameters of each component abnormal event are used to construct a binary scatter plot of the server cluster component abnormal event operation parameters;

[0091] Substitute the binary scatter plot of the abnormal event operation parameters of the server cluster component into the abnormal event classification model of the server cluster component, perform cyclic iteration to divide the optimal distance between each scatter point in the binary scatter plot and the decision surface, take the minimization error function of the decision boundary graph of the model training as the end goal, and generate the abnormal event data distribution decision boundary graph of the server cluster component;

[0092] The server cluster component abnormal event classification model is specifically:

[0093]

[0094] In the formula, F ij is the decision boundary diagram of abnormal event data distribution of server cluster components, w * is the optimal weight vector, b * is the optimal bias term, is the optimal value of the Lagrange multiplier associated with the kth abnormal operation characteristic parameter in the binary scatter plot, y k is the true label of the kth abnormal operation feature parameter in the binary scatter plot, x kis the kth abnormal operation characteristic parameter in the binary scatter plot, y c is the true label of the cth support vector, x c is the eigenvector of the c-th support vector;

[0095] It can be understood that the above model construction process solves the objective function and constraints through the Lagrangian algorithm and the kernel function. However, in order to ensure that the abnormal operation characteristic parameters corresponding to the abnormal events of the server cluster components can be correctly divided into positive and negative samples, soft intervals are used for the intervals between support vectors. This is a method well known to practitioners in this field and will not be described in detail here.

[0096] This solution builds a server cluster component abnormal event classification model, learns the server cluster component abnormal event operating parameters, and generates a decision boundary diagram for the abnormal event data distribution of the server cluster component. Based on this, it provides a reference object for real-time anomaly detection of server cluster components and improves the detection efficiency of abnormal events of server cluster components.

[0097] Reference Figure 5 As shown in the figure, based on the abnormal event data distribution decision boundary diagram of the server cluster component and the real-time operation data of the server cluster component, the abnormal event development direction of the server cluster component is generated, which specifically includes:

[0098] Standardize the real-time running data of the components of the server cluster, associate it with the real-time running data of the components of the server cluster according to the unit time, and create the real-time running sequence data of the components of the server cluster;

[0099] Normalize the real-time component health status values ​​of the server cluster and assign health status weights to the corresponding components of the server cluster.

[0100] Based on the component health status weights of the server cluster, the real-time running time data of the server cluster components, and the abnormal event data distribution decision boundary diagram of the server cluster components, the initial server cluster component abnormal development direction matching value is obtained;

[0101] Taking the unit time as the recursive step length, the matching growth value of the abnormal development direction of the initial server cluster components under the cumulative recursive step length is analyzed;

[0102] Determine whether the matching growth value of the initial server cluster component abnormal development direction under the cumulative recursive step length is within the abnormal event matching threshold range of the server cluster component; if not, continue to observe; if so, issue a corresponding server cluster component abnormal event;

[0103] The matching value of the abnormal development direction of the initial server cluster component is specifically:

[0104]

[0105] In the formula, cos i is the abnormal event matching value of the i-th component of the server cluster, Δcos i Match the growth value of the abnormal event of the i-th component of the server cluster, cos i (t+1) is the matching growth value of the abnormal development direction of the initial server cluster component under the cumulative t+1th recursive step, Q i is the real-time operating parameter of the i-th component of the server cluster, w i is the health status weight of the i-th component of the server cluster.

[0106] It can be understood that the purpose of assigning health status weights to the corresponding components of the server cluster is to more accurately reflect the impact of the current health status of each component on the overall system stability when calculating the matching value of the abnormal development direction, thereby optimizing the accuracy and timeliness of abnormality detection;

[0107] This solution standardizes the real-time operation data of server cluster components and associates and matches it with the decision boundary diagram of the abnormal event data distribution to achieve real-time monitoring and early warning of potential abnormal events, thereby improving the accuracy of anomaly detection of server cluster components. At the same time, by analyzing the matching growth value of the cumulative recursive step length, the development trend of abnormal events is visualized.

[0108] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions only describe the principles of the present invention. The present invention may be subject to various changes and improvements without departing from the spirit and scope of the present invention. These changes and improvements fall within the scope of the present invention. The scope of protection claimed by the present invention is defined by the attached claims and their equivalents.

Claims

1. A server cluster anomaly diagnosis method based on big data AI, characterized in that: include: Based on the historical maintenance logs of the server cluster, the health growth value of each component of the server cluster due to the historical maintenance task is quantified to obtain the component maintenance health growth coefficient of the server cluster. Based on the historical operation logs of the server cluster, analyze the deviation trend of the historical operation data of each component of the server cluster and evaluate the component initialization health indicators of the server cluster; According to the component maintenance health growth coefficient of the server cluster, the component initialization health index of the server cluster is corrected to evaluate the real-time component health status of the server cluster; Collect real-time operation data of components in the server cluster according to the real-time component health status of the server cluster; Capture abnormal events of server cluster components based on the Internet, obtain operational characteristic parameters of abnormal events of server cluster components, and construct an operational characteristic parameter matrix of abnormal events of server cluster components; Based on the abnormal event operation characteristic parameter matrix of the server cluster component, a classification model of the abnormal event operation parameters of the server cluster component is constructed, and a decision boundary diagram of the abnormal event data distribution of the server cluster component is generated; Based on the abnormal event data distribution decision boundary diagram of the server cluster components and the real-time operation data of the server cluster components, the development direction of the abnormal events of the server cluster components is generated.

2. According to the server cluster abnormality diagnosis method based on big data AI in claim 1, it is characterized in that: Based on the historical maintenance logs of the server cluster, the health growth value of each component of the server cluster due to the historical maintenance task is quantified. The component maintenance health growth coefficient of the server cluster is evaluated, including: Based on the historical maintenance log of the server cluster, extract the historical maintenance record of the server cluster component in unit time; the historical maintenance record includes: maintenance time, maintenance type, and maintenance component; Divide the historical maintenance records of the server cluster components according to unit time to obtain a time series data array of the maintenance records of the server cluster components; According to the maintenance record time series data of the server cluster component, the performance value of the maintenance component corresponding to the start time node of the maintenance task is used as the initial health benchmark value, and the maintenance type of the maintenance task for the health growth value of the maintenance component is calculated; Count the average growth value of the health growth value of the maintenance component for all maintenance types of the maintenance tasks in the maintenance record time series data of the server cluster component; The component maintenance health growth coefficient of the server cluster is calculated based on the health growth value of the maintenance type of the maintenance task for the maintenance component and the average growth value of the health growth value of the maintenance type of the maintenance task for the maintenance component.

3. According to the server cluster abnormality diagnosis method based on big data AI according to claim 2, it is characterized in that: The component maintenance health growth coefficient of the server cluster is specifically: Where Hup i Maintain a healthy growth coefficient, ΔH, for the i-th component of the server cluster ij is the health growth value of the i-th component of the server cluster after the j-th maintenance task, is the average health growth value of the i-th component of the server cluster, is the average health growth value of all components in the server cluster, n i is the total maintenance times of the i-th component, are the performance score of the i-th component of the server cluster after the j-th maintenance task, and the performance score of the i-th component of the server cluster before the j-th maintenance task, respectively. are the maximum performance score of the i-th component of the server cluster in the j-th maintenance task, the minimum performance score of the i-th component of the server cluster in the j-th maintenance task, and W ij The jth maintenance task weight for the i-th component of the server cluster.

4. According to the server cluster abnormality diagnosis method based on big data AI in claim 3, it is characterized in that: Based on the historical operation logs of the server cluster, analyze the deviation trend of the historical operation data of each component of the server cluster, and evaluate the component initialization health indicators of the server cluster, including: Based on the historical operation logs of the server cluster, the operation data of the server cluster components are associated according to the time attributes of the historical operation logs to create the server cluster component operation sequence data; Preprocessing is performed based on the server cluster component runtime data, and the operation characteristic parameters of each component in the server cluster component runtime data are counted, and an operation deviation threshold T is set; the operation deviation threshold is a preset constant, which is obtained based on the normal distribution Z score of the data; The time span between the maintenance time nodes of the historical maintenance tasks is used as the sliding window length, and the abnormal operation characteristic parameters of each component operation characteristic parameter within the sliding window length that exceeds the operation deviation threshold are recursively filtered out. Among them, e′ i is the abnormal operation characteristic parameter of the i-th component, e i is the operating characteristic parameter of the i-th component, is the mean value of the operating characteristic parameter, S is the standard deviation of the operating characteristic parameter; Based on the abnormal operation characteristic parameters of each component of the server cluster within the sliding window length, mark the abnormal deviation degree and abnormal deviation time duration under the corresponding abnormal state, and calculate the abnormal operation deviation weight of each component; the abnormal deviation degree refers to the absolute value between the abnormal operation characteristic parameter and the normal operation characteristic parameter; Based on the abnormal operation characteristic parameters of each component of the server cluster and the abnormal operation deviation weight of each component, the component initialization health index of the server cluster is calculated.

5. According to the method of server cluster abnormality diagnosis based on big data AI in claim 4, it is characterized in that: The component initialization health indicators of the server cluster are specifically: In the formula, G i Initialize the health indicator, W′, for the i-th component of the server cluster ik is the kth abnormal operation feature parameter weight of the i-th component of the server cluster, D ik is the deviation value of the kth abnormal operation characteristic parameter of the i-th component of the server cluster, U ik is the deviation time of the kth abnormal operation characteristic parameter of the i-th component of the server cluster, W′ max is the maximum weight of the abnormal operation characteristic parameter of the server cluster component within the sliding window length, and m is the total number of abnormal operation characteristic parameters.

6. A server cluster abnormality diagnosis method based on big data AI according to claim 5, characterized in that: Based on the abnormal event operation parameter matrix of the server cluster component, a classification model of the abnormal event operation parameters of the server cluster component is constructed, and a decision boundary diagram of the abnormal event data distribution of the server cluster component is generated. Specifically, the following are included: Based on the principal component analysis method, the dimension reduction process is performed on each element in the abnormal event operation parameter matrix of the server cluster component to obtain the reduced dimension server cluster component abnormal event operation parameter matrix; Based on SVM vector machine, a classification model of abnormal events of server cluster components is constructed; Based on the dimension-reduced server cluster component abnormal event operation parameter matrix, the dimension-reduced operation parameters of each component abnormal event are used to construct a binary scatter plot of the server cluster component abnormal event operation parameters; The binary scatter plot of the operating parameters of abnormal events of server cluster components is substituted into the abnormal event classification model of server cluster components, and the optimal distance between each scatter point in the binary scatter plot and the decision surface is divided cyclically and iteratively. The minimization of the error function of the decision boundary graph of model training is taken as the end goal, and the decision boundary graph of the abnormal event data distribution of the server cluster components is generated.

7. The server cluster abnormality diagnosis method based on big data AI according to claim 6 is characterized in that: The server cluster component abnormal event classification model is specifically: In the formula, F ij is the decision boundary diagram of abnormal event data distribution of server cluster components, w * is the optimal weight vector, b * is the optimal bias term, is the optimal value of the Lagrange multiplier associated with the kth abnormal operation characteristic parameter in the binary scatter plot, y k is the true label of the kth abnormal operation feature parameter in the binary scatter plot, x k is the kth abnormal operation characteristic parameter in the binary scatter plot, y c is the true label of the cth support vector, x c is the eigenvector of the cth support vector.

8. The server cluster abnormality diagnosis method based on big data AI according to claim 7 is characterized in that: Based on the abnormal event data distribution decision boundary diagram of the server cluster component and the real-time operation data of the server cluster component, the abnormal event development direction of the server cluster component is generated, which specifically includes: Standardize the real-time running data of the components of the server cluster, associate it with the real-time running data of the components of the server cluster according to the unit time, and create the real-time running sequence data of the components of the server cluster; Normalize the real-time component health status values ​​of the server cluster and assign health status weights to the corresponding components of the server cluster. Based on the component health status weights of the server cluster, the real-time running time data of the server cluster components, and the abnormal event data distribution decision boundary diagram of the server cluster components, the initial server cluster component abnormal development direction matching value is obtained; Taking the unit time as the recursive step length, the matching growth value of the abnormal development direction of the initial server cluster components under the cumulative recursive step length is analyzed; Determine whether the initial server cluster component abnormal development direction matching growth value under the cumulative recursive step is within the abnormal event matching threshold range of the server cluster component. If not, continue to observe. If so, issue the corresponding server cluster component abnormal event.

9. The server cluster abnormality diagnosis method based on big data AI according to claim 8 is characterized in that: The specific matching value of the abnormal development direction of the initial server cluster component is: In the formula, cos i is the abnormal event matching value of the i-th component of the server cluster, Δcos i Match the growth value of the abnormal event of the i-th component of the server cluster, cos i (t+1) is the matching growth value of the abnormal development direction of the initial server cluster component under the cumulative t+1th recursive step, Q i is the real-time operating parameter of the i-th component of the server cluster, w i is the health status weight of the i-th component of the server cluster.

Images